[Planetlab-users] On RTT measurement accuracy

[Scott Atchley]

Hi Soshant,

What are you trying to determine using the RTT metric? Something about 
the underlying topology or how applications will behave when 
communicating over the network? I would not argue against using 
application layer RTT if you are interested in the latter case. Just 
because the application layer takes longer than the ICMP ping, my 
application does not communicate via ICMP, but via application layer 
sockets which will be impacted by the overhead of the slices, etc.

Scott

On Jul 29, 2004, at 10:59 PM, Soshant Bali wrote:

> Hi,
>  
> Click on the link below to see the graph of RTTs collected over the 
> last few days between nodes
>  planetlab-2.scla.nodes.planet-lab.org  and 
> planet1.seattle.intel-research.net.
>  
> http://www.ittc.ku.edu/~sbali/PLabRTTs
>  
> Clearly, the average RTT on July 28 and 29 is higher than the average 
> RTT prior to that. A similar abrupt increase in average RTT on July 28 
> is also observed on other PlanetLab nodes where our program was 
> running. (The new CPU scheduling was deployed on July 27-28).
>  
> These RTTs were collected using a simple UDP echo client-server 
> program. Since packets are timestamped at the application layer in 
> such a setup, my guess is that the packets are getting delayed in the 
> client/server node while waiting for other slices to relinquish 
> control of CPU.
>  
> But PlanetLab nodes still seem to be very responsive to ICMP ping. The 
> link below has some RTTs collected on July 29 between the same two 
> nodes using the standard ping instead of echo client-server.
>  
> http://www.ittc.ku.edu/~sbali/ping
>  
> Clearly the high RTTs are absent in this data set.
>   
> So, to conclude, overlay, CDN and other services running on PlanetLab 
> should use ICMP ping and not some application layer echo 
> client-server to measure RTTs, now that the CPU sharing mechanism has 
> been deployed.
>  
> Is this correct? Are other people observing similar high delays?
>  
> Thanks
> Soshant
>  
>  
> ----- Original Message -----
>  From: "Steve Muir" <smuir at cs.princeton.edu>
> To: <users at lists.planet-lab.org>
> Sent: Tuesday, July 27, 2004 4:24 PM
> Subject: [Planetlab-users] enabling CPU shares on PlanetLab nodes
>
> > further to Marc F's note last week, we've accelerated the deployment 
> of
> > CPU scheduling in order to try and prevent CPU hogs taking over 
> nodes in
> > the days prior to the upcoming WORLDS and HotNets deadlines: this 
> change
>  > will now be deployed to all production nodes tonight.  we've been 
> running
> > with shares enabled on beta nodes for several weeks now and have not 
> seen
> > any problems, please let us know if you encounter any odd behaviour.
> > note that the required upgrade of plkmod will also cause all nodes to
>  > reboot as they download the update.
> >
> > cheers,
> >
> > steve
> >
> > _______________________________________________
> > Users mailing list: Users at lists.planet-lab.org
> > http://lists.planet-lab.org/mailman/listinfo/users
> _______________________________________________
> Users mailing list: Users at lists.planet-lab.org
> http://lists.planet-lab.org/mailman/listinfo/users